The introduction of OpenAI’s ChatGPT on November 30, 2022, marked a pivotal moment in the public’s engagement with artificial intelligence. Initially presented as a research preview, this conversational system rapidly captivated millions, offering a tangible, interactive experience with generative AI. Users across the globe tested its capabilities, challenging its limits, praising its fluency, and sharing screenshots of its often-impressive outputs. What began as an experimental curiosity has, in a remarkably short span of three and a half years, transformed into an indispensable part of digital infrastructure, seamlessly integrated into search engines, office productivity suites, customer-service platforms, coding environments, and corporate workflows. This profound shift, however, has unveiled a critical and emerging challenge: agency decay.
Dr. Cornelia C. Walther, a distinguished visiting scholar at Wharton and director of the global alliance POZE, has brought this phenomenon to the forefront of her research. With over two decades of experience as a humanitarian practitioner at the United Nations, Dr. Walther’s current work focuses on leveraging AI for social good, making her insights into the potential downsides of unexamined AI integration particularly salient. She defines agency decay as the gradual erosion of an individual’s fundamental capacity and inclination to observe meticulously, think independently, make deliberate choices, and act with responsibility. Crucially, this decay is not an inherent flaw of AI itself, but rather a byproduct of convenience becoming the default mode of cognition. A tool designed to augment human thought can, insidiously, begin to think around us, then for us, and eventually, without us even recognizing the subtle weakening of our own cognitive faculties.
The Genesis of Agency Decay: From Novelty to Necessity
The rapid ascent of generative AI, exemplified by ChatGPT, has been characterized by an unprecedented rate of adoption. What began as a niche technological marvel quickly permeated mainstream consciousness. Early adopters, often tech enthusiasts and innovators, were quick to explore its potential, showcasing its ability to draft coherent text, summarize complex documents, and even generate creative content. This initial "wow" factor fueled a widespread belief in AI’s transformative power, leading to a scramble among businesses and individuals alike to integrate these tools into their daily operations. The promise of enhanced productivity, reduced workload, and accelerated processes proved irresistible, accelerating the journey along what Dr. Walther terms the "Scale of Agency Decay." This progression, from initial experimentation to full dependence, has outpaced the development of adequate governance frameworks, training protocols, and cultural adjustments necessary to mitigate its risks. Every individual and organization navigating the modern digital landscape is, to varying degrees, somewhere along this continuum.
Stage One: The Era of Experimentation and Early Gains
The initial phase of AI adoption is largely characterized by experimentation. Users engage with generative AI tools in a playful and pragmatic manner, leveraging them for tasks such as drafting emails, summarizing lengthy articles, generating creative titles, translating text, explaining complex concepts, debugging code, and producing preliminary document versions. The business case for this immediate utility was compelling. Early research provided concrete evidence supporting the excitement; a significant study published in The Quarterly Journal of Economics, for instance, revealed that access to a generative AI assistant boosted productivity among customer-support agents by an average of 15%. This improvement was most pronounced among less experienced and lower-skilled workers, demonstrating AI’s potential to democratize certain skills—enabling a junior employee to communicate more clearly, a non-native speaker to write more fluently, or an overwhelmed manager to consolidate disparate notes into actionable prose.
At this nascent stage, human oversight remains firmly in place. The AI functions as an enhancer, expanding the reach and efficiency of human effort. However, even here, a subtle risk emerges: the early, tangible gains can foster a broad, uncritical assumption that faster output invariably equates to better work. This assumption proves fragile, particularly when tasks demand nuanced judgment, contextual understanding, ethical consideration, or direct accountability—qualities that AI, in its current form, struggles to genuinely replicate.
Stage Two: Integrating AI into Core Workflows – The Shift from Tool to Collaborator
The transition to the second stage, integration, marks a fundamental shift in the relationship between humans and AI. Generative AI moves beyond being an occasional side experiment, becoming a pervasive layer embedded within everyday work processes. This deeper integration fundamentally alters how tasks are approached and decisions are made. A growing body of research illustrates this shift, indicating that workers are not merely using AI more frequently, but are actively incorporating its outputs directly into their core drafts, analyses, and recommendations. In essence, AI systems begin to be treated less as mere tools and more as genuine collaborators.
Within enterprise environments, AI is now deeply embedded across various functions, from customer service chatbots acting as a first point of contact to advanced systems aiding in legal research, financial analysis, and software development. This embedding is often accompanied by an increase in user trust. Studies have shown that individuals frequently defer to AI-generated suggestions, even when these conflict with their own initial judgment, particularly if the system has performed well on prior tasks. An intriguing related finding suggests a correlation between lower trust in human counterparts and a higher propensity to trust AI.
This is where the perilous transition from "cognitive offloading" to "belief offloading" commences. Cognitive offloading, wherein AI handles routine or memory-intensive tasks, might seem innocuous. However, as integration deepens, individuals begin to outsource not just mental effort but also critical evaluation. Research into automation bias and "algorithm appreciation" reveals a concerning trend: individuals are more likely to accept AI-generated answers as correct, even when presented with contradictory evidence. The central question subtly shifts from "Is this task suitable for AI?" to "What did the AI say?" This seemingly minor change dramatically alters the locus of judgment, moving it from human intellect to algorithmic output.
The real-world consequences of this unchecked integration are becoming increasingly evident and problematic. In the notable case of Mata v. Avianca Inc., lawyers faced sanctions after submitting legal filings that contained fabricated cases, a direct output of ChatGPT. Similarly, in Canada, the Moffatt v. Air Canada decision held the airline accountable after its chatbot provided a customer with inaccurate information regarding bereavement fares. These incidents underscore a broader pattern: integrated AI systems are often trusted beyond their actual reliability, and the crucial step of human verification is gradually supplanted by uncritical acceptance. The initial promise of augmented human capability risks devolving into a system where human critical thinking is bypassed, with potentially severe legal, ethical, and operational repercussions.
Stage Three: The Peril of Reliance – Unexamined Trust and "Workslop"
Reliance marks the stage where AI becomes the default starting point for many tasks. While individuals still retain the responsibility to review the output, the initial generative impulse has irrevocably shifted from human ideation to machine generation. This pervasive reliance is clearly reflected in recent adoption statistics. Data from the Pew Research Center indicated that by 2025, 34% of U.S. adults had used ChatGPT, a figure that effectively doubled since 2023. Among employed adults, a significant 28% reported using it for work-related tasks. Further reinforcing this trend, Gallup’s third-quarter 2025 survey found that 45% of U.S. employees utilized AI at work at least a few times a year, with 23% engaging with it weekly and 10% on a daily basis.
The issue is not merely the frequency of AI usage, but the unexamined nature of this reliance. A 2025 Microsoft Research study, involving 319 knowledge workers and analyzing 936 instances of generative AI use in the workplace, yielded a stark finding: higher confidence in AI correlated with a demonstrable reduction in critical thinking. Conversely, greater confidence in one’s own task ability was associated with more critical thinking. The uncomfortable truth is simple: the more users trust the AI tool, the more conscientiously they must cultivate and protect the habit of questioning its outputs.
This uncritical reliance is now generating tangible costs within organizations. BetterUp Labs, in collaboration with the Stanford Social Media Lab, has coined the term "workslop" to describe AI-generated work that appears polished but fundamentally lacks substance, effectively transferring the burden of genuine thought onto colleagues. Their survey of 1,150 full-time U.S. desk workers revealed that 40% had encountered workslop in the preceding month. Each such incident required approximately two hours to resolve, translating to an estimated monthly cost of $186 per employee. This phenomenon represents agency decay at a team level: one individual saves time by outsourcing cognitive effort, while another subsequently loses time attempting to reconstruct the missing intellectual rigor. The cumulative effect is a decline in overall organizational quality, innovation, and efficiency.
Stage Four: Dependence – The Apex of Erosion and Systemic Vulnerabilities
Dependence represents the most advanced stage of agency decay, characterized by individuals or teams struggling significantly to perform, make decisions, or verify information without the aid of AI. While this stage is still theoretically reversible, the warning signs are increasingly visible and concerning. A 2025 study from the MIT Media Lab, which examined LLM-assisted essay writing using electroencephalography (EEG) to measure brain activity, offered a profound insight. Participants who used large language models exhibited weaker neural connectivity, reported a diminished sense of ownership over their essays, and found it more challenging to recall and quote their own work compared to those who wrote without AI assistance. While a subsequent scholarly comment raised valid concerns regarding sample size, reproducibility, and aspects of the EEG analysis, the study nevertheless poses a critical question for educational institutions, employers, and leaders globally: what are the long-term implications for skill formation and cognitive development when the foundational process of drafting and ideation is increasingly outsourced?
At the enterprise level, dependence manifests as "scale without depth." McKinsey’s 2026 global survey on AI revealed that a staggering 88% of organizations were deploying AI in at least one business function. Yet, many were still in the nascent stages of scaling AI effectively and realizing enterprise-level value. The same survey concurrently reported a range of negative consequences stemming from AI use, with inaccuracy emerging as one of the leading issues. This suggests that while organizations are embracing AI for speed and breadth, they may be sacrificing depth of understanding, robust verification, and genuine innovation, creating brittle systems that are highly susceptible to AI’s inherent limitations.
Beyond individual and organizational impacts, dependence on AI carries a significant planetary dimension. At the point of use, AI often feels immaterial, an invisible digital force. However, its operation is anything but. The International Energy Agency (IEA) projects a dramatic increase in electricity demand from data centers, potentially more than doubling by 2030, with AI being a primary driver, according to its analysis of energy demand from AI. This escalating energy consumption has profound environmental implications, contributing to carbon emissions and straining existing power grids. Furthermore, the U.S. Government Accountability Office (GAO), in its report on generative AI’s environmental and human effects, highlighted that generative AI consumes substantial energy and water resources. Worryingly, companies frequently fail to provide sufficient detail regarding these impacts, hindering proper accountability. Critically, there are currently no binding legal obligations compelling these companies to be transparent about their environmental footprint. Thus, safeguarding human agency amidst AI now extends beyond protecting our individual attention and cognitive faculties; it necessitates a holistic approach that links personal convenience with institutional design and the broader planetary costs of unchecked technological adoption.
Reversing the Tide: The A-Frame for Deliberate Co-Agency
While agency decay is a tangible and growing concern, it is, for now, a reversible trend. The overarching objective must be to transition from passive AI adoption to deliberate co-agency—a symbiotic relationship where humans and AI collaborate in ways that actively strengthen human judgment, cultivate essential skills, foster responsibility, and instill a profound care for the wider systems upon which life depends. Dr. Walther proposes the A-Frame as a practical starting point for this critical reversal.
Awareness: The journey begins with heightened awareness, specifically by noticing the precise moment of delegation to AI. Before engaging an AI tool, individuals should pause and critically ask: "What exactly am I entrusting to this machine?" Is it merely drafting, or is it reasoning, judgment, empathy, verification, or even accountability? On an organizational level, teams must systematically map where AI is introduced into workflows and rigorously identify points where direct human review and intervention are not just advisable, but absolutely essential. This proactive assessment ensures that AI is used intentionally, rather than as a default.
Appreciation: True appreciation means judiciously deploying AI where it genuinely expands and augments human capacity. AI excels at tasks such as summarizing vast amounts of information, translating languages, comparing complex datasets, running simulations, and revealing blind spots that human cognition might miss. However, the distinctly human work of framing the problem, establishing core values, understanding nuanced context, and defining what "good" or "ethical" truly means must be preserved and championed. AI should be a powerful co-pilot, not an autonomous navigator. This stage emphasizes understanding AI’s strengths while simultaneously recognizing and valuing the irreplaceable human contributions.
Acceptance: This pillar demands unvarnished honesty about AI’s inherent limitations. Despite its sophisticated outputs, AI can sound utterly certain and yet be entirely incorrect. Similarly, individuals can experience a superficial sense of productivity while simultaneously witnessing their underlying skills atrophy. Organizations, in their rush to innovate, can scale AI tools far in advance of their ability to scale the critical human judgment required to oversee them. Consequently, every AI-enabled workflow must be meticulously designed to incorporate robust verification standards, clear escalation paths for anomalies, and regular opportunities for individuals to practice the underlying skills without automation. This ensures that human competence remains sharp and responsive, preventing over-reliance.
Accountability: The final pillar, accountability, transforms individual agency into robust organizational governance. On a personal level, cultivating a "human first draft" habit for any task requiring significant critical thinking or original ideation is paramount. This ensures the initial cognitive heavy lifting is performed by the human mind. Within teams, it is crucial to mandate source checks for AI-generated information, transparently disclose meaningful AI use in projects, and explicitly assign human owners for all AI-supported decisions. Furthermore, organizational procurement and strategy for AI must explicitly consider and account for the energy, water, and infrastructure impacts of these technologies. This comprehensive approach ensures that the benefits of AI are weighed against its full spectrum of costs, fostering responsible and sustainable innovation.
The Future of Work: Agency as a Strategic Asset
In the evolving landscape of artificial intelligence, the organizations that choose to treat human agency not as a quaint relic but as a vital business asset will be the ones that thrive. Concepts such as individual autonomy, nuanced judgment, earned trust, and unwavering planetary responsibility are not merely "soft" concerns; they represent the foundational infrastructure for sustainable performance and long-term viability. Companies that proactively protect and cultivate these human attributes will be uniquely positioned to leverage AI with greater discipline, imagination, and, critically, legitimacy. The choice is clear: succumb to the subtle erosion of agency, or deliberately champion human-AI co-agency, forging a future where technology empowers humanity without diminishing its essence.
About the Author
